Agent Security Gap Puts Enterprises at Risk

A significant security gap exists as organizations rapidly deploy AI agents without adequate controls, according to new research. Over half (54%) of enterprises have already experienced an agent security incident or near-miss, while fundamental security practices lag behind.

Key Findings

• Incident rates are high: 18% confirmed incidents, 36% near misses
• Identity management is weak: Only 32% give every agent its own identity; most share credentials
• Isolation measures insufficient: Just 30% isolate their highest-risk agents
• Security stacks are borrowed: Organizations primarily use controls from model providers (OpenAI, Google Cloud, etc.) rather than purpose-built agent security solutions
• Satisfaction masks risk: While enterprises rate existing controls highly (4.2/5), most acknowledge defenses aren’t keeping pace with AI threats and plan to change tooling soon